infrastructure/stack.yml (1)

143-143: Route Grafana over TLS (websecure) and keep HTTP->HTTPS redirect.

Networks look correct for Traefik (public) and Prometheus (internal). Recommend switching the Grafana router to websecure and enabling TLS, plus an HTTP redirect to harden access.

Example labels to add (no diff since lines are outside the changed hunk):
yaml
labels:
- traefik.enable=true
- traefik.http.routers.grafana.rule=Host(grafana.example.com)
- traefik.http.routers.grafana.entrypoints=websecure
- traefik.http.routers.grafana.tls=true
- traefik.http.routers.grafana-web.rule=Host(grafana.example.com)
- traefik.http.routers.grafana-web.entrypoints=web
- traefik.http.routers.grafana-web.middlewares=redirect-to-https
- traefik.http.middlewares.redirect-to-https.redirectscheme.scheme=https

libraries/python/src/internal/tft_model.py (1)

124-124: Remove unused noqa / dead comment line.

Ruff flags this (RUF100). Delete the commented-out line rather than carrying a suppressed noqa.

-        # NOTE: static_continuous_features = Tensor.zeros(self.batch_size, 1, 0)  # noqa: E501, ERA001

libraries/python/src/internal/company_information.py (1)

15-26: Consider vectorized checks for performance and clearer errors.

Element-wise Python lambdas can be slower. If feasible, use column-scoped checks or vectorized predicates for each column to produce column-specific errors.
Example (conceptual, adjust to pandera-polars API):
python
checks=[
pa.Check(lambda df: df["sector"].str.to_uppercase() == df["sector"], element_wise=True, error="sector must be uppercase"),
pa.Check(lambda df: df["industry"].str.to_uppercase() == df["industry"], element_wise=True, error="industry must be uppercase"),
pa.Check(lambda df: df["sector"].str.strip_chars() == df["sector"], element_wise=True, error="sector must be stripped"),
pa.Check(lambda df: df["industry"].str.strip_chars() == df["industry"], element_wise=True, error="industry must be stripped"),
]

If you prefer to keep top-level checks, add a test covering missing values to ensure the new uppercase defaults behave as intended.

libraries/python/src/internal/prediction.py (2)

16-18: Remove unused noqa comments flagged by Ruff (RUF100).

The # noqa: E501 markers are unnecessary here.

Apply:

-        message = f"Each ticker must have exactly {dates_count} unique dates, found: {unique_dates_per_ticker}"  # noqa: E501
+        message = f"Each ticker must have exactly {dates_count} unique dates, found: {unique_dates_per_ticker}"
@@
-                message = f"Expected all tickers to have the same dates, mismatch between: {first_ticker_dates} and: {set(dates)}"  # noqa: E501
+                message = f"Expected all tickers to have the same dates, mismatch between: {first_ticker_dates} and: {set(dates)}"

Also applies to: 34-35

---

40-75: Optional: enforce non-empty frame if required by contract.

Empty inputs currently pass both cross-record checks. If at least one ticker/week is required, add a schema-level check for non-empty.

libraries/python/tests/test_prediction.py (2)

19-19: Ruff S101 in tests — OK to ignore or disable for test paths.

If Ruff gates CI, add a per-file or dir-level ignore for S101 in tests.

Example pyproject fragment (outside this file):

[tool.ruff]
extend-exclude = []

[tool.ruff.per-file-ignores]
"**/tests/**/*.py" = ["S101"]

Also applies to: 79-79, 143-143

---

7-17: Optional: add coercion tests.

Since prediction_schema has coerce=True, add a case with string timestamps/quantiles to assert dtype coercion occurs.

Also applies to: 131-144

applications/portfoliomanager/src/portfoliomanager/alpaca_client.py (4)

60-61: Avoid double validation of positions.

You validate in get_account and again in AlpacaAccount.__init__. Keep one (prefer the entity constructor to enforce invariants).

Apply:

-        position_schema.validate(positions)
+        # rely on AlpacaAccount to validate on construction

Also applies to: 17-19

---

27-27: Remove unused noqa (RUF100).

# noqa: FBT001 is unnecessary.

Apply:

-        is_paper: bool,  # noqa: FBT001
+        is_paper: bool,

---

62-65: Optional: use Decimal for cash to avoid float drift.

Financial amounts benefit from exact arithmetic. Consider Decimal in AlpacaAccount and callers.

---

74-96: Schema consistency with existing helper.

Nice reuse of is_uppercase shape. Consider importing the shared helper from internal.portfolio to avoid duplication if both exist.

applications/portfoliomanager/tests/test_risk_management.py (2)

31-31: Remove unused noqa directives

These noqa directives are not needed as the corresponding rules are not enabled in your linting configuration.

-    assert len(result) == 2  # noqa: PLR2004
+    assert len(result) == 2

Apply similar changes to lines 51, 107, 430-432, 436, 467, 498, 525, 561, 571, 575, 611, and 625.

Also applies to: 51-51, 107-107, 430-432, 436-436, 467-467, 498-498, 525-525, 561-561, 571-571, 575-575, 611-611, 625-625

---

427-427: Consider using UTC timezone consistently

While datetime.now(tz=UTC) works, consider using a fixed timestamp for test reproducibility.

-        predictions, positions, 20000.0, datetime.now(tz=UTC)
+        predictions, positions, 20000.0, datetime(2024, 1, 15, tzinfo=UTC)

This suggestion applies to all occurrences of datetime.now(tz=UTC) in the test file.

libraries/python/src/internal/tft_dataset.py (1)

29-31: Consider documenting the API change rationale

The constructor signature change from __init__(self, data: pl.DataFrame) to __init__(self) followed by preprocess_and_set_data(data: pl.DataFrame) is a significant API change. Consider adding a docstring explaining the two-step initialization pattern.

    def __init__(self) -> None:
+        """Initialize TFTDataset without data.
+        
+        Data should be provided via preprocess_and_set_data() method to allow
+        for better separation of initialization and data processing steps.
+        """
        pass

libraries/python/tests/test_tft_dataset.py (1)

188-244: Redundant assertion on Line 244.

Line 240 already checks that targets.shape[1] == 1, making the identical check on Line 244 redundant.

             targets = batch["targets"]
             assert targets.shape[1] == 1
             expected_tensor_dimensions = 2
             assert len(targets.shape) == expected_tensor_dimensions
             assert targets.shape[0] > 0
-            assert targets.shape[1] == 1

applications/portfoliomanager/src/portfoliomanager/risk_management.py (3)

216-224: Consider making the uncertainty threshold configurable.

The hardcoded minimum_inter_quartile_range = 0.75 might need adjustment based on market conditions or strategy requirements.

 def create_optimal_portfolio(
     predictions: pl.DataFrame,
     positions: pl.DataFrame,
     maximum_capital: float,
     current_timestamp: datetime,
+    minimum_inter_quartile_range: float = 0.75,
 ) -> pl.DataFrame:
     predictions = predictions.clone()
     positions = positions.clone()
 
-    minimum_inter_quartile_range = 0.75
     high_uncertainty_tickers = (

---

261-262: Consider making position count targets configurable.

The hardcoded target of 10 positions per side might need adjustment based on portfolio size or risk tolerance.

 def create_optimal_portfolio(
     predictions: pl.DataFrame,
     positions: pl.DataFrame,
     maximum_capital: float,
     current_timestamp: datetime,
+    target_positions_per_side: int = 10,
 ) -> pl.DataFrame:

Then update lines 261-262:

-    new_long_positions_needed = max(0, 10 - maintained_long_count)
-    new_short_positions_needed = max(0, 10 - maintained_short_count)
+    new_long_positions_needed = max(0, target_positions_per_side - maintained_long_count)
+    new_short_positions_needed = max(0, target_positions_per_side - maintained_short_count)

---

23-24: Remove unused noqa directives.

Multiple noqa directives for FBT003 and E501 are unused according to static analysis. These should be removed to keep the code clean.

Remove the # noqa: FBT003 and # noqa: E501 comments from lines 23, 24, 43, 153, 154, 173, 174, 201, 202, 203, 287, 288, 296, and 297, as they're not suppressing any actual linting errors.

Also applies to: 43-43, 153-154, 173-174, 201-203, 287-288, 296-297

libraries/python/src/internal/portfolio.py (3)

30-31: Remove unused noqa comments flagged by Ruff.

E501 isn’t enabled; drop the # noqa: E501 to satisfy RUF100.

-        message = f"Expected {total_positions_count} total positions, found: {data.lazyframe.collect().height}"  # noqa: E501
+        message = f"Expected {total_positions_count} total positions, found: {data.lazyframe.collect().height}"

-        message = f"Expected long and short dollar amount sums to be within {maximum_imbalance_percentage * 100}%, found long: {long_sum}, short: {short_sum}"  # noqa: E501
+        message = f"Expected long and short dollar amount sums to be within {maximum_imbalance_percentage * 100}%, found long: {long_sum}, short: {short_sum}"

Also applies to: 57-58

---

88-96: Pass check functions directly; avoid unnecessary lambdas.

Cleaner and keeps signatures discoverable.

-        pa.Check(
-            check_fn=lambda df: check_position_side_counts(df),
-            error="Each side must have expected position counts",
-        ),
-        pa.Check(
-            check_fn=lambda df: check_position_side_sums(df),
-            error="Position side sums must be approximately equal",
-        ),
+        pa.Check(
+            check_fn=check_position_side_counts,
+            error="Each side must have expected position counts",
+        ),
+        pa.Check(
+            check_fn=check_position_side_sums,
+            error="Position side sums must be approximately equal",
+        ),

---

65-83: Consider making columns non-nullable.

If nulls are not expected, enforce nullable=False to catch missing data early.

-        "ticker": pa.Column(
-            dtype=str,
-            checks=[pa.Check(is_uppercase)],
-        ),
+        "ticker": pa.Column(
+            dtype=str,
+            checks=[pa.Check(is_uppercase)],
+            nullable=False,
+        ),
...
-        "timestamp": pa.Column(
-            dtype=int,
-            checks=[pa.Check.greater_than(0)],
-        ),
+        "timestamp": pa.Column(
+            dtype=int,
+            checks=[pa.Check.greater_than(0)],
+            nullable=False,
+        ),
...
-        "side": pa.Column(
-            dtype=str,
-            checks=[
+        "side": pa.Column(
+            dtype=str,
+            checks=[
                 pa.Check.isin(["LONG", "SHORT"]),
                 pa.Check(is_uppercase),
-            ],
-        ),
+            ],
+            nullable=False,
+        ),
...
-        "dollar_amount": pa.Column(
-            dtype=float,
-            checks=[pa.Check.greater_than(0)],
-        ),
+        "dollar_amount": pa.Column(
+            dtype=float,
+            checks=[pa.Check.greater_than(0)],
+            nullable=False,
+        ),

libraries/python/tests/test_portfolio.py (3)

84-117: Assert error-message substrings are brittle; prefer tighter matching or parametrization.

Substring matches can pass unexpectedly. Consider parametrizing tests to cover both long/short mismatch cases and assert full messages.

-    with pytest.raises(SchemaError, match="Expected 10 long side positions, found: 15"):
+    with pytest.raises(
+        SchemaError,
+        match=r"^Expected 10 long side positions, found: 15$",
+    ):
         portfolio_schema.validate(data)

Optionally, parametrize LONG/SHORT imbalances to reduce duplication.

Also applies to: 119-153

---

38-40: Ruff S101 in tests.

If Ruff flags assert in tests, add a per-file ignore or disable S101 for tests in ruff config.

Example ruff.toml snippet:

[tool.ruff]
extend-exclude = ["libraries/python/tests/**"]

[tool.ruff.per-file-ignores]
"libraries/python/tests/**" = ["S101"]

---

10-36: Reduce boilerplate ticker lists.

Use a helper/fixture to generate N distinct tickers to keep tests short and focused.

-            "ticker": [
-                "AAPL", "GOOGL", "MSFT", "AMZN", "TSLA", "NVDA", "META", "NFLX", "BABA", "CRM",
-                "AMD", "INTC", "ORCL", "ADBE", "PYPL", "SHOP", "SPOT", "ROKU", "ZM", "DOCU",
-            ],
+            "ticker": [f"T{i:03d}" for i in range(20)],

Also applies to: 122-148

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

libraries/python/src/internal/mhsa_network.py (1)

39-39: Remove unused noqa (RUF100).

ERA001 isn’t enabled; drop the directive.

-        # shape: (batch, heads_count, sequence_length, head_dimension)  # noqa: ERA001
+        # shape: (batch, heads_count, sequence_length, head_dimension)

maskfile.md (2)

434-436: Fix positional parameter parsing in test_endpoint.

"$1/$2/$3" are required; current values are literals.

-    local name="1$"
-    local url="2$"
-    local context="3$"
+    local name="$1"
+    local url="$2"
+    local context="$3"

---

472-473: Fix PortfolioManager port in production tests.

Use 8081 to match the local URLs and service mapping.

-    test_endpoint "PortfolioManager" "http://$MANAGER_IP:808/health" "production"
-    test_endpoint "PortfolioManager (root)" "http://$MANAGER_IP:808/" "production"
+    test_endpoint "PortfolioManager" "http://$MANAGER_IP:8081/health" "production"
+    test_endpoint "PortfolioManager (root)" "http://$MANAGER_IP:8081/" "production"

applications/portfoliomanager/src/portfoliomanager/risk_management.py (2)

96-107: Log-return guard looks good; resolves prior concern.

---

372-377: Narrow exception handling and (optionally) log.

Catching Exception hides real issues; catch the expected types.

-    except Exception:  # noqa: BLE001
-        return 0.0
+    except (TypeError, ValueError):
+        # logger.warning("Error calculating capital for side %s", side)
+        return 0.0

applications/portfoliomanager/src/portfoliomanager/alpaca_client.py (1)

49-49: Apply the previous review fix for side parsing

-                "side": str(account_position.side).replace("PositionSide.", "").upper(),
+                "side": str(account_position.side).replace("PositionSide.", "").upper(),

Note: This appears to already be correctly implemented, so this is just confirming the fix was applied.

applications/portfoliomanager/pyproject.toml (1)

12-12: Pin upper bound for alpaca-py 0.x series.

0.x minors can be breaking; bound to <0.43 to avoid surprise upgrades.

-    "alpaca-py>=0.42.1",
+    "alpaca-py>=0.42.1,<0.43",

libraries/python/tests/test_company_information.py (2)

101-103: Defaults for None → "NOT AVAILABLE" look good; add a combined-null case

Nice pivot from exception-based behavior to defaults. Consider adding a test where both fields are None to lock in the intended behavior.

Example to add outside this hunk:

def test_company_information_schema_both_null_defaults() -> None:
    data = pl.DataFrame({"sector": [None], "industry": [None]})
    df = company_information_schema.validate(data)
    assert df["sector"][0] == "NOT AVAILABLE"
    assert df["industry"][0] == "NOT AVAILABLE"

---

101-103: Silence Ruff/Bandit S101 on asserts in tests (or ignore in config)

Ruff flags S101 here. If your pipeline enforces it on tests, either add per-line ignores or configure per-file ignores.

Per-line option (minimal change):

-    assert validated_df["sector"][0] == "NOT AVAILABLE"
-    assert validated_df["industry"][0] == "SOFTWARE"
+    assert validated_df["sector"][0] == "NOT AVAILABLE"  # noqa: S101
+    assert validated_df["industry"][0] == "SOFTWARE"  # noqa: S101

-    assert validated_df["industry"][0] == "NOT AVAILABLE"
-    assert validated_df["sector"][0] == "TECHNOLOGY"
+    assert validated_df["industry"][0] == "NOT AVAILABLE"  # noqa: S101
+    assert validated_df["sector"][0] == "TECHNOLOGY"  # noqa: S101

Config option (outside this file, preferred):

# pyproject.toml
[tool.ruff.lint.per-file-ignores]
"libraries/python/tests/test_company_information.py" = ["S101"]

Also applies to: 114-116

libraries/python/src/internal/tft_model.py (1)

124-124: Remove unused noqa (RUF100).

No need to suppress E501/ERA001 on a commented-out line.

-        # NOTE: static_continuous_features = Tensor.zeros(self.batch_size, 1, 0)  # noqa: E501, ERA001
+        # NOTE: static_continuous_features = Tensor.zeros(self.batch_size, 1, 0)

libraries/python/src/internal/prediction.py (3)

6-9: Reuse shared is_uppercase helper to avoid duplication.

Import the existing helper from internal.portfolio for consistency.

-import pandera.polars as pa
-import polars as pl
-from pandera.polars import PolarsData
+import pandera.polars as pa
+import polars as pl
+from pandera.polars import PolarsData
+from internal.portfolio import is_uppercase
@@
-def is_uppercase(data: PolarsData) -> pl.LazyFrame:
-    return data.lazyframe.select(
-        pl.col(data.key).str.to_uppercase() == pl.col(data.key)
-    )
+# reused from internal.portfolio

---

23-24: Drop unused E501 no-qa tags (RUF100).

Long f-strings don’t need suppression if you keep them; or wrap onto multiple lines.

-        message = f"Each ticker must have exactly {dates_count} unique dates, found: {unique_dates_per_ticker}"  # noqa: E501
+        message = f"Each ticker must have exactly {dates_count} unique dates, found: {unique_dates_per_ticker}"
@@
-                message = f"Expected all tickers to have the same dates, mismatch between: {first_ticker_dates} and: {set(dates)}"  # noqa: E501
+                message = f"Expected all tickers to have the same dates, mismatch between: {first_ticker_dates} and: {set(dates)}"

Also applies to: 40-41

---

62-71: Align Check.error messages with raised ValueErrors (keeps tests stable).

Pandera may surface either message; make them identical to the exceptions you raise.

-        pa.Check(
-            check_fn=lambda df: check_dates_count_per_ticker(df),
-            name="check_dates_count_per_ticker",
-            error="Each ticker must have expected date count",
-        ),
+        pa.Check(
+            check_fn=lambda df: check_dates_count_per_ticker(df),
+            name="check_dates_count_per_ticker",
+            error="Each ticker must have exactly 7 unique dates",
+        ),
@@
-        pa.Check(
-            check_fn=lambda df: check_same_dates_per_ticker(df),
-            name="check_same_dates_per_ticker",
-            error="All tickers must have same date values",
-        ),
+        pa.Check(
+            check_fn=lambda df: check_same_dates_per_ticker(df),
+            name="check_same_dates_per_ticker",
+            error="Expected all tickers to have the same dates",
+        ),

maskfile.md (2)

198-198: Use a non-zero SSH ConnectTimeout.

Zero causes immediate failure; use something reasonable for retries.

-    if ssh -o ConnectTimeout=0 pocketsizefund-production 'docker info -f "{{.ServerVersion}} {{.Swarm.LocalNodeState}}"' 2>/dev/null; then
+    if ssh -o ConnectTimeout=10 pocketsizefund-production 'docker info -f "{{.ServerVersion}} {{.Swarm.LocalNodeState}}"' 2>/dev/null; then

---

218-220: Standardize redirections.

Prefer >/dev/null 2>&1 when you intend to suppress both stdout and stderr.

Example:

-docker context use default >/dev/null >&1 || true
+docker context use default >/dev/null 2>&1 || true

Apply similarly across the listed lines.

Also applies to: 229-231, 341-341, 412-412, 479-479, 493-493, 511-511, 686-686, 692-692, 697-697

libraries/python/src/internal/tft_dataset.py (3)

130-149: Zero-fill vs. >0 constraints conflict. Choose one strategy.

You fill missing OHLC with 0.0, then enforce >0 in schema—this will fail whenever data were missing. Either:

• allow zeros in schema (>= 0), or
• avoid zero-fill for OHLC (forward-fill/backfill or drop rows) and keep >0.

Option A (permit zero placeholders):

-        "open_price": pa.Column(dtype=float, checks=pa.Check.greater_than(0)),
+        "open_price": pa.Column(dtype=float, checks=pa.Check.greater_than_or_equal_to(0)),
-        "high_price": pa.Column(dtype=float, checks=pa.Check.greater_than(0)),
+        "high_price": pa.Column(dtype=float, checks=pa.Check.greater_than_or_equal_to(0)),
-        "low_price": pa.Column(dtype=float, checks=pa.Check.greater_than(0)),
+        "low_price": pa.Column(dtype=float, checks=pa.Check.greater_than_or_equal_to(0)),
-        "close_price": pa.Column(dtype=float, checks=pa.Check.greater_than(0)),
+        "close_price": pa.Column(dtype=float, checks=pa.Check.greater_than_or_equal_to(0)),

Option B (preferred for data quality): replace the 0.0 fills for OHLC with a forward-fill/backfill per ticker and keep the >0 checks. I can draft that if you prefer.

Also applies to: 406-420

---

229-233: Use consistent Polars expressions for min/max dates.

Mirror earlier style for clarity and to avoid mixing Series/Python values in select.

-        minimum_date: date = self.data.select(self.data["date"].min()).item()
-        maximum_date: date = self.data.select(self.data["date"].max()).item()
+        minimum_date: date = self.data.select(pl.col("date").min()).item()
+        maximum_date: date = self.data.select(pl.col("date").max()).item()

---

341-347: Fix type hint for input_batch.

The function expects the batch dict, not a single Tensor.

-    def postprocess_predictions(
-        self,
-        input_batch: Tensor,  # static_categorical_features
+    def postprocess_predictions(
+        self,
+        input_batch: dict[str, Tensor],  # expects batch dict

libraries/python/tests/test_tft_dataset.py (3)

118-126: Make batch count assertion less brittle.

Future changes to windowing could validly yield >1 batch. Prefer ≥1.

-    assert len(batches) == 1
+    assert len(batches) >= 1

---

238-245: Remove duplicate assertion.

targets.shape[1] == 1 is asserted twice.

-            assert targets.shape[1] == 1
             expected_tensor_dimensions = 2
             assert len(targets.shape) == expected_tensor_dimensions
             assert targets.shape[0] > 0
             assert targets.shape[1] == 1

---

144-186: Ignore Bandit S101 for test asserts (or configure Ruff).

The S101 warnings are expected in pytest-style tests. Either keep as-is or add a test-only ignore in Ruff.

Add to pyproject.toml:

[tool.ruff.lint.per-file-ignores]
"libraries/python/tests/**" = ["S101"]

libraries/python/src/internal/equity_bar.py (1)

46-50: Nit: comment typo.

“partion” → “partition”.

-    strict="filter",  # allows DuckDB partion columns
+    strict="filter",  # allows DuckDB partition columns

applications/portfoliomanager/src/portfoliomanager/risk_management.py (6)

13-26: Vectorize PDT lock derivation; drop map_elements.

Leverage Polars datetime casting for speed and clarity.

-    return positions.with_columns(
-        pl.when(
-            pl.col("timestamp")
-            .cast(pl.Int64)
-            .map_elements(
-                lambda ts: datetime.fromtimestamp(ts, tz=UTC).date(),
-                return_dtype=pl.Date,
-            )
-            == current_date
-        )
-        .then(True)  # noqa: FBT003
-        .otherwise(False)  # noqa: FBT003
-        .alias("pdt_locked")
-    )
+    return positions.with_columns(
+        (
+            pl.col("timestamp")
+            .cast(pl.Int64)
+            .cast(pl.Datetime(time_unit="s", time_zone="UTC"))
+            .dt.date()
+            == pl.lit(current_date)
+        ).alias("pdt_locked")
+    )

---

111-135: Replace Python loop with vectorized cumulative returns for scalability.

Iterating rows will bottleneck with many positions. Consider precomputing per-ticker cumulative sums and using asof joins.

I can provide a vectorized Polars rewrite if you want it in this PR.

---

115-121: Use integer seconds once and reuse.

Avoid float comparisons and repeated conversions.

-    for row in position_predictions.iter_rows(named=True):
+    current_ts_s = int(current_timestamp.timestamp())
+    for row in position_predictions.iter_rows(named=True):
@@
-            & (pl.col("timestamp") <= current_timestamp.timestamp())
+            & (pl.col("timestamp") <= current_ts_s)

---

139-181: Null thresholds produce null predicates; default to False explicitly.

Fill missing thresholds to avoid null-tristate booleans leaking into when/then.

-    return positions_with_data.with_columns(
+    return positions_with_data.with_columns(
+        pl.col("original_lower_threshold").fill_null(float("inf")).alias("original_lower_threshold"),
+        pl.col("original_upper_threshold").fill_null(float("-inf")).alias("original_upper_threshold"),

---

222-246: Ensure ordering before head/tail selection.

If callers forget to sort predictions via add_predictions_zscore_ranked_columns, selection becomes arbitrary. Sort defensively here.

-    available_predictions = predictions.filter(
-        ~pl.col("ticker").is_in(excluded_tickers)
-    )
+    available_predictions = (
+        predictions.filter(~pl.col("ticker").is_in(excluded_tickers))
+        .sort(["composite_score", "inter_quartile_range"], descending=[True, False])
+    )

---

23-24: Remove unused noqa FBT003 directives.

Rule FBT003 isn’t enabled; the directives create noise.

-        .then(True)  # noqa: FBT003
-        .otherwise(False)  # noqa: FBT003
+        .then(True)
+        .otherwise(False)

(Apply similarly to other instances in this file.)

Also applies to: 43-43, 159-160, 179-180, 207-209, 293-294, 302-303

applications/portfoliomanager/src/portfoliomanager/alpaca_client.py (2)

27-27: Remove unused noqa directive

The noqa: FBT001 directive is not needed since FBT001 is not enabled.

-        is_paper: bool,  # noqa: FBT001
+        is_paper: bool,

---

68-71: Code duplication: is_uppercase function exists elsewhere

The is_uppercase function is duplicated between this file and libraries/python/src/internal/portfolio.py. Consider importing it from the internal library to reduce duplication.

+from internal.portfolio import is_uppercase
-
-def is_uppercase(data: pa.PolarsData) -> pl.LazyFrame:
-    return data.lazyframe.select(
-        pl.col(data.key).str.to_uppercase() == pl.col(data.key)
-    )

applications/portfoliomanager/tests/test_risk_management.py (3)

89-91: Remove unused noqa directives; keep asserts in tests by configuring Ruff.

Ruff flags unused “noqa: E501/PLR2004” here. Drop them and, if needed, ignore S101 for tests in Ruff config.

Minimal cleanup examples:

-def test_add_equity_bars_returns_and_realized_volatility_columns_sufficient_data_success() -> (  # noqa: E501
-    None
-):
+def test_add_equity_bars_returns_and_realized_volatility_columns_sufficient_data_success() -> None:
@@
-    assert len(result) == 35  # noqa: PLR2004
+    assert len(result) == 35

Optional Ruff config (root pyproject.toml):

[tool.ruff.lint.per-file-ignores]
"**/tests/**" = ["S101"]

Also applies to: 110-112, 127-129, 148-150, 430-436, 467-470, 498-499, 525-527, 561-575, 611-625

---

406-433: Add negative-path test for empty optimal portfolio.

When all predictions are excluded and there are no maintained positions, create_optimal_portfolio should raise ValueError. Add a test to lock this behavior.

+def test_create_optimal_portfolio_raises_when_no_components() -> None:
+    predictions = pl.DataFrame({"ticker": ["HIGH"], "composite_score": [1.0], "inter_quartile_range": [0.9]})
+    positions = pl.DataFrame({"ticker": [], "timestamp": [], "side": [], "dollar_amount": [], "close_position": [], "maintain_position": []})
+    with pytest.raises(ValueError, match="No portfolio components"):
+        create_optimal_portfolio(predictions, positions, 20000.0, datetime.now(tz=UTC))

---

473-499: Strengthen assertions: verify 1 long and 1 short after high-uncertainty exclusion.

This makes the intent explicit and guards against future ordering changes.

 result = create_optimal_portfolio(
     predictions, positions, 20000.0, datetime.now(tz=UTC)
 )
 
 assert "HIGH_UNCERT" not in result["ticker"].to_list()
-assert len(result) == 2  # only 2 available predictions
+assert len(result) == 2
+assert result.filter(pl.col("side") == "LONG").height == 1
+assert result.filter(pl.col("side") == "SHORT").height == 1

applications/portfoliomanager/pyproject.toml (1)

5-5: Optional: relax exact Python patch pin.

Using a range reduces churn in CI while staying within 3.12.

-requires-python = "==3.12.10"
+requires-python = ">=3.12,<3.13"

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

libraries/python/src/internal/mhsa_network.py (1)

39-39: Remove unused noqa directive (RUF100).

Drop the # noqa: ERA001 from the shape comment or enable ERA001 in Ruff config. The directive is currently unused.

-        # shape: (batch, heads_count, sequence_length, head_dimension)  # noqa: ERA001
+        # shape: (batch, heads_count, sequence_length, head_dimension)

applications/portfoliomanager/src/portfoliomanager/risk_management.py (1)

373-386: Avoid broad except Exception; handle nulls directly.

Broad catches can hide real issues; summation can be expressed null-safe.

Apply:

     if filtered_positions.height == 0:
         return 0.0
 
-    try:
-        side_capital_amount = filtered_positions.select(pl.sum("dollar_amount")).item()
-        return float(side_capital_amount or 0)
-
-    except Exception:  # noqa: BLE001
-        return 0.0
+    side_capital_amount = filtered_positions.select(
+        pl.col("dollar_amount").sum()
+    ).item()
+    return float(side_capital_amount or 0.0)

libraries/python/src/internal/prediction.py (1)

45-51: Fix Polars uppercase check (use project helper or Polars string API).

Current lambda uses .upper() which is invalid for Polars Series in pandera-polars contexts; reuse the existing helper for consistency.

+from internal.portfolio import is_uppercase
@@
-        "ticker": pa.Column(
-            dtype=str,
-            checks=[
-                pa.Check(
-                    lambda s: s.upper() == s,
-                    error="Ticker must be uppercase",
-                    element_wise=True,
-                )
-            ],
-        ),
+        "ticker": pa.Column(
+            dtype=str,
+            checks=[pa.Check(is_uppercase, error="Ticker must be uppercase")],
+        ),

maskfile.md (2)

472-473: Correct PortfolioManager production port (808 → 8081).

Endpoints currently probe the wrong port.

-    test_endpoint "PortfolioManager" "http://$MANAGER_IP:808/health" "production"
-    test_endpoint "PortfolioManager (root)" "http://$MANAGER_IP:808/" "production"
+    test_endpoint "PortfolioManager" "http://$MANAGER_IP:8081/health" "production"
+    test_endpoint "PortfolioManager (root)" "http://$MANAGER_IP:8081/" "production"

---

434-449: Fix function parameters (placeholders printed literally today).

Positional args are malformed, causing incorrect output and statuses.

-test_endpoint() {
-    local name="1$"
-    local url="2$"
-    local context="3$"
+test_endpoint() {
+    local name="$1"
+    local url="$2"
+    local context="$3"

libraries/python/src/internal/tft_model.py (1)

124-124: Remove unused noqa directives on a commented-out line (RUF100).

The NOTE line is commented and still carries # noqa: E501, ERA001. Either delete the line or drop the noqa suffixes.

-        # NOTE: static_continuous_features = Tensor.zeros(self.batch_size, 1, 0)  # noqa: E501, ERA001
+        # NOTE: static_continuous_features = Tensor.zeros(self.batch_size, 1, 0)

applications/portfoliomanager/src/portfoliomanager/alpaca_client.py (3)

27-27: Remove unused noqa (RUF100).

FBT001 isn’t enabled; drop the directive.

-        is_paper: bool,  # noqa: FBT001
+        is_paper: bool,

---

60-65: Avoid double validation; validate in one place.

You validate positions here and again in AlpacaAccount. Keep only one (recommend keeping it in AlpacaAccount).

-        position_schema.validate(positions)
-
         return AlpacaAccount(
             cash_amount=float(cast("str", account.cash)),
             positions=positions,
         )

---

68-96: Consider deduping validation helpers.

is_uppercase duplicates logic in libraries/python/src/internal/portfolio.py. If import boundaries allow, reuse a single shared helper to avoid drift.

libraries/python/src/internal/portfolio.py (3)

6-9: Uppercase check helper mirrors another module.

To reduce duplication, consider centralizing is_uppercase for reuse (or import from portfoliomanager if layering permits).

---

12-15: Hard-coded total position count (20) may be too rigid.

If live portfolios aren’t always 20 positions, parameterize this via schema context/config or move the count check outside the schema.

---

78-112: Schema options: consider strictness.

If you want to forbid unexpected columns in portfolio inputs, set strict=True. If not, current coerce=True and unique=["ticker"] look fine.

-portfolio_schema = pa.DataFrameSchema(
+portfolio_schema = pa.DataFrameSchema(
     {
       ...
     },
     unique=["ticker"],
     coerce=True,
+    # strict=True,  # optionally enforce no extra columns
     checks=[
       ...
     ],
 )

libraries/python/tests/test_portfolio.py (3)

115-116: Relax brittle error-message matching.

Exact string match on the counts makes tests fragile. Prefer a stable substring.

Apply:

-with pytest.raises(SchemaError, match="Expected 10 long side positions, found: 15"):
+with pytest.raises(SchemaError, match=r"Expected\s+10\s+long\s+side\s+positions"):

---

151-152: Relax brittle error-message matching (amount sums).

Match on a stable phrase rather than the full sentence.

Apply:

-with pytest.raises(SchemaError, match="long and short dollar amount sums"):
+with pytest.raises(SchemaError, match=r"long\s+and\s+short\s+dollar\s+amount\s+sums"):

---

39-39: Ruff S101 in tests: ignore for test files.

Asserts in tests are fine. Consider a per-file-ignores rule instead of changing assertions.

Add to pyproject.toml:

[tool.ruff.lint.per-file-ignores]
"**/tests/**" = ["S101"]

applications/portfoliomanager/src/portfoliomanager/risk_management.py (6)

93-107: Cast end timestamp once to Int; avoid float comparison and repeated calls.

Safer and a touch faster.

Apply:

-    original_equity_bars_with_returns = original_equity_bars.sort(
-        ["ticker", "timestamp"]
-    )
+    original_equity_bars_with_returns = original_equity_bars.sort(
+        ["ticker", "timestamp"]
+    )
+    end_ts = int(current_timestamp.timestamp())
@@
-        ticker_bars = original_equity_bars_with_returns.filter(
+        ticker_bars = original_equity_bars_with_returns.filter(
             (pl.col("ticker") == ticker)
             & (pl.col("timestamp") >= position_timestamp)
-            & (pl.col("timestamp") <= current_timestamp.timestamp())
+            & (pl.col("timestamp") <= end_ts)
         )

---

130-179: Optional: validate dependency column presence.

If callers pass raw bars, log_daily_returns is missing and the sum yields None; fail fast with a clear message.

Apply near Line 130 before the join:

+    if "log_daily_returns" not in original_equity_bars_with_returns.columns:
+        raise ValueError("original_equity_bars must include 'log_daily_returns'; run add_equity_bars_returns_and_realized_volatility_columns first.")

---

198-201: Guard against empty predictions.

Mean/std on empty frames can produce None; return early.

Apply:

 def add_predictions_zscore_ranked_columns(predictions: pl.DataFrame) -> pl.DataFrame:
     predictions = predictions.clone()
+    if predictions.height == 0:
+        return predictions.with_columns(
+            pl.lit(0.0).alias("z_score_return"),
+            pl.lit(0.0).alias("inter_quartile_range"),
+            pl.lit(0.0).alias("composite_score"),
+            pl.lit(False).alias("pdt_locked"),
+            pl.lit(False).alias("close_position"),
+            pl.lit(False).alias("maintain_position"),
+        )

---

245-253: Naming: threshold is a maximum, not minimum.

minimum_inter_quartile_range reads backwards; rename for clarity.

Apply:

-    minimum_inter_quartile_range = 0.75
+    max_inter_quartile_range = 0.75
@@
-            pl.col("inter_quartile_range") > minimum_inter_quartile_range
+            pl.col("inter_quartile_range") > max_inter_quartile_range

---

281-297: Round monetary amounts to cents.

Avoid fractional cents in allocations.

Apply:

-        pl.lit(dollar_amount_per_long).alias("dollar_amount"),
+        pl.lit(dollar_amount_per_long).round(2).alias("dollar_amount"),
@@
-        pl.lit(dollar_amount_per_short).alias("dollar_amount"),
+        pl.lit(dollar_amount_per_short).round(2).alias("dollar_amount"),

---

23-24: Remove unused # noqa directives flagged by Ruff (RUF100).

These rules aren’t enabled; drop the comments to keep lint clean.

Example edits (apply similarly to all occurrences):

-        .then(True)  # noqa: FBT003
-        .otherwise(False)  # noqa: FBT003
+        .then(True)
+        .otherwise(False)

Also applies to: 42-42, 150-151, 170-171, 198-200, 286-287, 295-296

applications/portfoliomanager/tests/test_risk_management.py (2)

31-31: Clean up unused # noqa comments and consider ignoring S101 in tests.

These noqa entries (PLR2004/E501) aren’t active; remove them. If S101 triggers in tests, prefer a per-file-ignores config.

Example:

-    assert len(result) == 2  # noqa: PLR2004
+    assert len(result) == 2

And in pyproject.toml:

[tool.ruff.lint.per-file-ignores]
"**/tests/**" = ["S101"]

Also applies to: 51-51, 74-74, 85-86, 89-89, 105-108, 111-111, 128-128, 195-195, 536-538, 542-542, 573-573, 604-604, 631-631, 667-667, 677-677, 681-681, 717-731, 734-734

---

532-535: Stabilize time-dependent tests (optional).

Using datetime.now(tz=UTC) is fine here, but fixed timestamps improve determinism.

Apply, e.g.:

-    result = create_optimal_portfolio(
-        predictions, positions, 20000.0, datetime.now(tz=UTC)
-    )
+    fixed_now = datetime(2024, 1, 20, tzinfo=UTC)
+    result = create_optimal_portfolio(predictions, positions, 20000.0, fixed_now)

Also applies to: 569-572, 599-602, 627-629, 661-663, 710-715

libraries/python/src/internal/prediction.py (2)

10-18: Prefer n_unique over unique().len for performance and clarity.

Aggregate counts directly; avoid materializing per-ticker lists in Python.

-    grouped = data.lazyframe.group_by("ticker").agg(
-        pl.col("timestamp").unique().alias("unique_dates")
-    )
-
-    unique_dates_per_ticker = grouped.collect()["unique_dates"].to_list()
-
-    if not all(len(dates) == dates_count for dates in unique_dates_per_ticker):
-        message = f"Each ticker must have exactly {dates_count} unique dates, found: {unique_dates_per_ticker}"  # noqa: E501
-        raise ValueError(message)
+    counts = data.lazyframe.group_by("ticker").agg(
+        pl.col("timestamp").n_unique().alias("n_unique")
+    ).collect()
+    invalid = counts.filter(pl.col("n_unique") != dates_count)
+    if invalid.height > 0:
+        raise ValueError(f"Each ticker must have exactly {dates_count} unique dates")

---

34-35: Remove unused noqa E501 directives.

Ruff flags these as unused; line lengths are fine now.

-        message = f"Each ticker must have exactly {dates_count} unique dates, found: {unique_dates_per_ticker}"  # noqa: E501
+        message = f"Each ticker must have exactly {dates_count} unique dates, found: {unique_dates_per_ticker}"
@@
-                message = f"Expected all tickers to have the same dates, mismatch between: {first_ticker_dates} and: {set(dates)}"  # noqa: E501
+                message = f"Expected all tickers to have the same dates, mismatch between: {first_ticker_dates} and: {set(dates)}"

Also applies to: 17-18

libraries/python/tests/test_prediction.py (1)

131-143: Add a test for quantile monotonicity (if you adopt the new check).

Covers the essential constraint that distributions are coherent.

 def test_prediction_schema_float_quantile_values() -> None:
@@
     validated_df = prediction_schema.validate(valid_data)
     assert validated_df.shape == (7, 5)
+
+
+def test_prediction_schema_quantiles_not_monotonic_fails() -> None:
+    data = pl.DataFrame(
+        {
+            "ticker": ["AAPL"] * 7,
+            "timestamp": [1, 2, 3, 4, 5, 6, 7],
+            "quantile_10": [100.0] * 7,
+            "quantile_50": [150.0] * 7,
+            "quantile_90": [140.0] * 7,  # lower than q50 should fail
+        }
+    )
+    with pytest.raises(SchemaError, match="Quantiles must be monotonic"):
+        prediction_schema.validate(data)

maskfile.md (3)

198-205: Use a non-zero SSH ConnectTimeout.

With 0s, every attempt fails immediately despite retries.

-    if ssh -o ConnectTimeout=0 pocketsizefund-production 'docker info -f "{{.ServerVersion}} {{.Swarm.LocalNodeState}}"' 2>/dev/null; then
+    if ssh -o ConnectTimeout=10 pocketsizefund-production 'docker info -f "{{.ServerVersion}} {{.Swarm.LocalNodeState}}"' 2>/dev/null; then

---

636-641: Show stack tasks (don’t discard stdout).

The intent is to display tasks; current redirection hides them.

-echo "️ Infrastructure Stack:"
-docker stack ps infrastructure >/dev/null || echo "Infrastructure stack not deployed"
+echo "️ Infrastructure Stack:"
+if ! docker stack ps infrastructure; then
+  echo "Infrastructure stack not deployed"
+fi
@@
-echo " Applications Stack:"
-docker stack ps applications >/dev/null || echo "Applications stack not deployed"
+echo " Applications Stack:"
+if ! docker stack ps applications; then
+  echo "Applications stack not deployed"
+fi

---

495-499: Actually print service list in health check.

Don’t redirect the table to /dev/null if you want operators to see it.

-        if docker service ls --format "table {{.Name}}\t{{.Replicas}}\t{{.Image}}" >/dev/null; then
-            echo " Services listed successfully"
+        if docker service ls --format "table {{.Name}}\t{{.Replicas}}\t{{.Image}}"; then
+            :
         else
             echo "  No services found or connection error"
         fi

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro